
import uuid
import json
import os

from tqdm.notebook import tqdm

import cv2


%pylab inline
import pandas as pd
from sklearn import preprocessing

Populating the interactive namespace from numpy and matplotlib


%load_ext autoreload
%autoreload 2

# configurations pandas
pd.set_option('display.max_rows', 250)
pd.set_option('display.max_columns', 250)


path = './data/Flipkart'
data = pd.read_csv(os.path.join(path,'flipkart_com-ecommerce_sample_1050.csv'))
print(data.shape)

data.drop([
    'crawl_timestamp', 'product_url', 'pid', 'retail_price', 'discounted_price', 'is_FK_Advantage_product', 'product_rating',
    'overall_rating','brand', 'product_category_1', 'product_category_2', 'product_category_3', 'product_category_4',
    'product_category_5', 'product_category_6'
], axis=1, inplace=True, errors='ignore')

def product_specification_decode(json_str):
    s = json_str.lower().replace("=>", ':').replace('{"product_specification":nil}', '{"product_specification":""}')
    return json.loads(s)

def product_category_tree_decode(json_str):
    cat_list = json.loads(json_str.lower())
    assert(isinstance(cat_list, list))
    assert(len(cat_list) == 1)
    return [s.strip() for s in cat_list[0].split('>>')]

data['uniq_id'] = data['uniq_id'].apply(lambda x: uuid.UUID(x))
data['product_category_tree'] = data['product_category_tree'].apply(product_category_tree_decode)
data[[f"product_category_{i}" for i in range(7)]] = pd.DataFrame(data['product_category_tree'].tolist(), index= data.index)
data['description'] = data['description'].apply(lambda x: x.lower())
data['product_name'] = data['product_name'].apply(lambda x: x.lower())
data['product_name_description'] = data['product_name'] + ' ' + data['description']
data.set_index('uniq_id', inplace=True, verify_integrity=True)

# Label encoding pour cat 0
le = preprocessing.LabelEncoder()
product_category_0 = data['product_category_0'].unique()
le.fit(product_category_0)
data['product_category_0_label'] = le.transform(data['product_category_0'])
data.head()

(1050, 15)


sample_data = data
#sample_data = data.sample(25)


# TEST
# Loading the image
img = cv2.imread('./data/Flipkart/Images/0a3b5fdf77a361c2d7d9b29c259b8c4e.jpg')

# Converting image to grayscale
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

# Applying SIFT detector
sift = cv2.SIFT_create()
kp, desc = sift.detectAndCompute(gray, None)
print("descriptor shape:", desc.shape)

# Marking the keypoint on the image using circles
cv2.drawKeypoints(gray, kp, img)

plt.figure(figsize=(30,15))
plt.imshow(img)

descriptor shape: (1558, 128)

<matplotlib.image.AxesImage at 0x23d14d61700>


# TEST
# no kp with this image
img = cv2.imread('./data/Flipkart/Images/9743e6d1c9f3180483825fe4d2e97836.jpg')

# Converting image to grayscale
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)

# Applying SIFT detector
sift = cv2.SIFT_create()
kp, desc = sift.detectAndCompute(gray, None)
print("sift keypoint:", len(kp))

sift keypoint: 0


# histogram
hist, bins = np.histogram(gray.flatten(),256,[0,256])

cdf = hist.cumsum()
cdf_normalized = cdf * float(hist.max()) / cdf.max()
plt.plot(cdf_normalized, color = 'b')
plt.hist(img.flatten(),256,[0,256], color = 'r')
plt.xlim([0,256])
plt.legend(('cdf','histogram'), loc = 'upper left')
plt.show()

plt.imshow(gray)
plt.show()

cdf_m = np.ma.masked_equal(cdf,0)
cdf_m = (cdf_m - cdf_m.min())*255/(cdf_m.max()-cdf_m.min())
cdf = np.ma.filled(cdf_m,0).astype('uint8')

gray2 = cdf[gray]

plt.imshow(gray2)
plt.show()


# TEST cv equalizeHist
equ = cv2.equalizeHist(gray)
res = np.hstack((gray, equ)) #stacking images side-by-side
plt.imshow(res)

<matplotlib.image.AxesImage at 0x23d15029970>


# create a CLAHE object (Arguments are optional).
# TODO filtre bruit
_img = img
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
cl1 = clahe.apply(gray)
plt.imshow(cl1)

# Applying SIFT detector
sift = cv2.SIFT_create()
kp, desc = sift.detectAndCompute(cl1, None)
print("sift keypoint:", len(kp))

# Marking the keypoint on the image using circles
cv2.drawKeypoints(gray, kp, _img)

plt.figure(figsize=(30,15))
plt.imshow(_img)

sift keypoint: 1

<matplotlib.image.AxesImage at 0x23d150e0580>


from src.opencv_app import SiftApp
app = SiftApp(sample_data['image'].to_dict(), sample_data['product_category_0'].to_dict())
app.create_embedding_matrix(with_filters=True)
app.create_kmean_model()
app.create_histogram()

Loading images...

number of images : 1050 , number of class : 7
Create sift wrapper...

Detecting keypoints...

Creating kmean model...
k = 70 , batch_size = 3150
Make histogram...


from sklearn.manifold import TSNE

_df, _ari = app.compute_ari(n_components=2, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[SIFT] Clustering des produits par TSNE 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[SIFT] Clustering des produits par TSNE 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

Computing ARI with TSNE(2d)...
ARI with TSNE(2): 0.0524
Plot clustering...

Plot images with categories...

Computing ARI with TSNE(3d)...
ARI with TSNE(3): 0.0436
Plot clustering...

Plot images with categories...


from sklearn.decomposition import PCA

_df, _ari = app.compute_ari(n_components=2, model=PCA)
app.plot_clustering(_df, title=f"[SIFT] Clustering des produits par PCA 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=PCA)
app.plot_clustering(_df, title=f"[SIFT] Clustering des produits par PCA 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

Computing ARI with PCA(2d)...
ARI with PCA(2): 0.0406
Plot clustering...

Plot images with categories...

Computing ARI with PCA(3d)...
ARI with PCA(3): 0.0518
Plot clustering...

Plot images with categories...


from tensorflow.keras.applications import VGG16
layers = VGG16().layers
for i in (0, -2, -1):
    print("name:", layers[i].name, "- shape:", layers[i].get_output_at(0).get_shape().as_list())

name: input_1 - shape: [None, 224, 224, 3]
name: fc2 - shape: [None, 4096]
name: predictions - shape: [None, 1000]


from src.keras_app import KerasApp, VggWrapper
app = KerasApp(sample_data['image'].to_dict(), sample_data['product_category_0'].to_dict(), model_wrapper_class=VggWrapper)
app.preprocess_images(target_size=(224, 224))

app.process_features(force_reload=True)
app.get_model_features().head()

from sklearn.manifold import TSNE

_df, _ari = app.compute_ari(n_components=2, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[CNN VGG16] Clustering des produits par TSNE 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[CNN VGG16] Clustering des produits par TSNE 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

from sklearn.decomposition import PCA

_df, _ari = app.compute_ari(n_components=2, model=PCA)
app.plot_clustering(_df, title=f"[CNN VGG16] Clustering des produits par PCA 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=PCA)
app.plot_clustering(_df, title=f"[CNN VGG16] Clustering des produits par PCA 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

Loading images...

number of images : 1050 , number of class : 7
Create VggWrapper wrapper...

Preprocess images...

c:\Users\flori\miniconda3\envs\jupyter\lib\site-packages\PIL\Image.py:2918: DecompressionBombWarning: Image size (93680328 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.
  warnings.warn(

Process features...

Computing ARI with TSNE(2d)...
ARI with TSNE(2): 0.4393
Plot clustering...

Plot images with categories...

Computing ARI with TSNE(3d)...
ARI with TSNE(3): 0.4543
Plot clustering...

Plot images with categories...

Computing ARI with PCA(2d)...
ARI with PCA(2): 0.2918
Plot clustering...


from tensorflow.keras.applications import InceptionV3
layers = InceptionV3().layers
for i in (0, -2, -1):
    print("name:", layers[i].name, "- shape:", layers[i].get_output_at(0).get_shape().as_list())

name: input_6 - shape: [None, 299, 299, 3]
name: avg_pool - shape: [None, 2048]
name: predictions - shape: [None, 1000]


from src.keras_app import KerasApp, InceptionWrapper
app = KerasApp(sample_data['image'].to_dict(), sample_data['product_category_0'].to_dict(), model_wrapper_class=InceptionWrapper)
app.preprocess_images(target_size=(299, 299))

app.process_features(force_reload=True)
app.get_model_features().head()

from sklearn.manifold import TSNE

_df, _ari = app.compute_ari(n_components=2, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[CNN InceptionV3] Clustering des produits par TSNE 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[CNN InceptionV3] Clustering des produits par TSNE 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

from sklearn.decomposition import PCA

_df, _ari = app.compute_ari(n_components=2, model=PCA)
app.plot_clustering(_df, title=f"[CNN InceptionV3] Clustering des produits par PCA 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=PCA)
app.plot_clustering(_df, title=f"[CNN InceptionV3] Clustering des produits par PCA 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

Loading images...

number of images : 1050 , number of class : 7
Create InceptionWrapper wrapper...

Preprocess images...

c:\Users\flori\miniconda3\envs\jupyter\lib\site-packages\PIL\Image.py:2918: DecompressionBombWarning: Image size (93680328 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.
  warnings.warn(

Process features...

Computing ARI with TSNE(2d)...
ARI with TSNE(2): 0.0591
Plot clustering...

Plot images with categories...

Computing ARI with TSNE(3d)...
ARI with TSNE(3): 0.0557
Plot clustering...

Plot images with categories...

Computing ARI with PCA(2d)...
ARI with PCA(2): 0.0347
Plot clustering...


from tensorflow.keras.applications import ResNet50V2
layers = ResNet50V2().layers
for i in (0, -2, -1):
    print("name:", layers[i].name, "- shape:", layers[i].get_output_at(0).get_shape().as_list())

name: input_7 - shape: [None, 224, 224, 3]
name: avg_pool - shape: [None, 2048]
name: predictions - shape: [None, 1000]


from src.keras_app import KerasApp, ResNetWrapper
app = KerasApp(sample_data['image'].to_dict(), sample_data['product_category_0'].to_dict(), model_wrapper_class=ResNetWrapper)
app.preprocess_images(target_size=(224, 224))

app.process_features(force_reload=True)
app.get_model_features().head()

from sklearn.manifold import TSNE

_df, _ari = app.compute_ari(n_components=2, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[CNN ResNet50V2] Clustering des produits par TSNE 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=TSNE, model_kw={'perplexity': 30, 'n_iter': 2000, 'init': 'random', 'learning_rate': 200, 'random_state': 42})
app.plot_clustering(_df, title=f"[CNN ResNet50V2] Clustering des produits par TSNE 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

from sklearn.decomposition import PCA

_df, _ari = app.compute_ari(n_components=2, model=PCA)
app.plot_clustering(_df, title=f"[CNN ResNet50V2] Clustering des produits par PCA 2D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

_df, _ari = app.compute_ari(n_components=3, model=PCA)
app.plot_clustering(_df, title=f"[CNN ResNet50V2] Clustering des produits par PCA 3D (ARI={_ari:.4f})")
app.plot_images(_df, n_sample=25)

Loading images...

number of images : 1050 , number of class : 7
Create ResNetWrapper wrapper...

Preprocess images...

c:\Users\flori\miniconda3\envs\jupyter\lib\site-packages\PIL\Image.py:2918: DecompressionBombWarning: Image size (93680328 pixels) exceeds limit of 89478485 pixels, could be decompression bomb DOS attack.
  warnings.warn(

Process features...

Computing ARI with TSNE(2d)...
ARI with TSNE(2): 0.0326
Plot clustering...

Plot images with categories...

Computing ARI with TSNE(3d)...
ARI with TSNE(3): 0.0307
Plot clustering...

Plot images with categories...

Computing ARI with PCA(2d)...
ARI with PCA(2): 0.0171
Plot clustering...


!jupyter nbconvert --config nbconvert/config_html.py

[NbConvertApp] Converting notebook P6_01_notebookexploration.ipynb to html
[NbConvertApp] Writing 787215 bytes to P6_01_notebookexploration.html
[NbConvertApp] Converting notebook P6_02_notebookbagofword.ipynb to html
[NbConvertApp] Writing 3641906 bytes to P6_02_notebookbagofword.html
[NbConvertApp] Converting notebook P6_03_notebookembedding.ipynb to html
[NbConvertApp] Writing 8347677 bytes to P6_03_notebookembedding.html
[NbConvertApp] Converting notebook P6_04_notebookimages.ipynb to html
[NbConvertApp] Writing 47617971 bytes to P6_04_notebookimages.html

	product_name	product_category_tree	image	description	product_specifications	product_category_0	product_category_1	product_category_2	product_category_3	product_category_4	product_category_5	product_category_6	product_name_description	product_category_0_label
uniq_id
55b85ea1-5a15-36d4-6b71-90ad6fff8ce7	elegance polyester multicolor abstract eyelet ...	[home furnishing, curtains & accessories, curt...	55b85ea15a1536d46b7190ad6fff8ce7.jpg	key features of elegance polyester multicolor ...	{"product_specification"=>[{"key"=>"Brand", "v...	home furnishing	curtains & accessories	curtains	elegance polyester multicolor abstract eyelet ...	None	None	None	elegance polyester multicolor abstract eyelet ...	4
7b72c92c-2f6c-4026-8628-ec5f14c6d590	sathiyas cotton bath towel	[baby care, baby bath & skin, baby bath towels...	7b72c92c2f6c40268628ec5f14c6d590.jpg	specifications of sathiyas cotton bath towel (...	{"product_specification"=>[{"key"=>"Machine Wa...	baby care	baby bath & skin	baby bath towels	sathiyas baby bath towels	sathiyas cotton bath towel (3 bath towel, red,...	None	None	sathiyas cotton bath towel specifications of s...	0
64d5d4a2-5824-3731-dc7b-bb1eef49ad74	eurospa cotton terry face towel set	[baby care, baby bath & skin, baby bath towels...	64d5d4a258243731dc7bbb1eef49ad74.jpg	key features of eurospa cotton terry face towe...	{"product_specification"=>[{"key"=>"Material",...	baby care	baby bath & skin	baby bath towels	eurospa baby bath towels	eurospa cotton terry face towel set (20 piece ...	None	None	eurospa cotton terry face towel set key featur...	0
d4684dcd-c759-dd9c-df41-504698d737d8	santosh royal fashion cotton printed king size...	[home furnishing, bed linen, bedsheets, santos...	d4684dcdc759dd9cdf41504698d737d8.jpg	key features of santosh royal fashion cotton p...	{"product_specification"=>[{"key"=>"Brand", "v...	home furnishing	bed linen	bedsheets	santosh royal fashion bedsheets	santosh royal fashion cotton printed king size...	None	None	santosh royal fashion cotton printed king size...	4
6325b687-0c54-cd47-be6e-bfbffa620ec7	jaipur print cotton floral king sized double b...	[home furnishing, bed linen, bedsheets, jaipur...	6325b6870c54cd47be6ebfbffa620ec7.jpg	key features of jaipur print cotton floral kin...	{"product_specification"=>[{"key"=>"Machine Wa...	home furnishing	bed linen	bedsheets	jaipur print bedsheets	jaipur print cotton floral king sized double b...	None	None	jaipur print cotton floral king sized double b...	4

Projet 6 : Classifiez automatiquement des biens de consommation¶

Notebook de prétraitement et exploration¶

Import¶

Configuration¶

Chargement des données¶

Extraction des features image avec l'algorithme : SIFT¶

Test du traitement des images¶

Clustering des images avec SIFT¶

Clustering des images avec CNN Transfer Learning¶

VGG16¶

Inception V3¶

ResNet50V2¶

Export du notebook en HTML¶